diff --git a/Data/Memory/ByteArray.hs b/Data/Memory/ByteArray.hs
new file mode 100644
index 0000000..19ae285
--- /dev/null
+++ b/Data/Memory/ByteArray.hs
@@ -0,0 +1,22 @@
+-- |
+-- Module      : Data.Memory.ByteArray
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : stable
+-- Portability : Good
+--
+-- Simple and efficient byte array types
+--
+{-# LANGUAGE BangPatterns #-}
+{-# LANGUAGE MagicHash #-}
+{-# LANGUAGE UnboxedTuples #-}
+{-# LANGUAGE NoImplicitPrelude #-}
+module Data.Memory.ByteArray
+    ( module X
+    ) where
+
+import           Data.Memory.ByteArray.Types         as X
+import           Data.Memory.ByteArray.Methods       as X
+import           Data.Memory.ByteArray.ScrubbedBytes as X (ScrubbedBytes)
+import           Data.Memory.ByteArray.Bytes         as X (Bytes)
+import           Data.Memory.ByteArray.MemView       as X (MemView(..))
diff --git a/Data/Memory/ByteArray/Bytes.hs b/Data/Memory/ByteArray/Bytes.hs
new file mode 100644
index 0000000..eaa5a35
--- /dev/null
+++ b/Data/Memory/ByteArray/Bytes.hs
@@ -0,0 +1,123 @@
+-- |
+-- Module      : Data.Memory.ByteArray.Bytes
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : stable
+-- Portability : Good
+--
+-- Simple and efficient byte array types
+--
+{-# LANGUAGE BangPatterns #-}
+{-# LANGUAGE MagicHash #-}
+{-# LANGUAGE UnboxedTuples #-}
+module Data.Memory.ByteArray.Bytes
+    ( Bytes
+    ) where
+
+import           GHC.Types
+import           GHC.Prim
+import           GHC.Ptr
+import           Data.Memory.Internal.CompatPrim
+import           Data.Memory.Internal.Compat      (unsafeDoIO)
+import           Data.Memory.ByteArray.Types
+import           Data.Memory.Encoding.Base16      (showHexadecimal)
+
+data Bytes = Bytes (MutableByteArray# RealWorld)
+
+instance Show Bytes where
+    show = bytesShowHex
+instance Eq Bytes where
+    (==) = bytesEq
+
+instance ByteArrayAccess Bytes where
+    length        = bytesLength
+    withByteArray = withBytes
+instance ByteArray Bytes where
+    allocRet = bytesAllocRet
+
+------------------------------------------------------------------------
+newBytes :: Int -> IO Bytes
+newBytes (I# sz) = IO $ \s ->
+    case newAlignedPinnedByteArray# sz 8# s of
+        (# s', mbarr #) -> (# s', Bytes mbarr #)
+
+touchBytes :: Bytes -> IO ()
+touchBytes (Bytes mba) = IO $ \s -> case touch# mba s of s' -> (# s', () #)
+
+sizeofBytes :: Bytes -> Int
+sizeofBytes (Bytes mba) = I# (sizeofMutableByteArray# mba)
+
+withPtr :: Bytes -> (Ptr p -> IO a) -> IO a
+withPtr b@(Bytes mba) f = do
+    a <- f (Ptr (byteArrayContents# (unsafeCoerce# mba)))
+    touchBytes b
+    return a
+------------------------------------------------------------------------
+
+{-
+bytesCopyAndModify :: Bytes -> (Ptr a -> IO ()) -> IO Bytes
+bytesCopyAndModify src f = do
+    dst <- newBytes sz
+    withPtr dst $ \d -> do
+        withPtr src $ \s -> copyBytes (castPtr d) s sz
+        f d
+    return dst
+  where sz = sizeofBytes src
+
+bytesTemporary :: Int -> (Ptr p -> IO a) -> IO a
+bytesTemporary sz f = newBytes sz >>= \ba -> withPtr ba f
+
+bytesCopyTemporary :: Bytes -> (Ptr p -> IO a) -> IO a
+bytesCopyTemporary src f = do
+    dst <- newBytes (sizeofBytes src)
+    withPtr dst $ \d -> do
+        withPtr src $ \s -> copyBytes (castPtr d) s (sizeofBytes src)
+        f d
+bytesAlloc :: Int -> (Ptr p -> IO ()) -> IO Bytes
+bytesAlloc sz f = do
+    ba <- newBytes sz
+    withPtr ba f
+    return ba
+-}
+
+bytesAllocRet :: Int -> (Ptr p -> IO a) -> IO (a, Bytes)
+bytesAllocRet sz f = do
+    ba <- newBytes sz
+    r <- withPtr ba f
+    return (r, ba)
+
+bytesLength :: Bytes -> Int
+bytesLength = sizeofBytes
+
+withBytes :: Bytes -> (Ptr p -> IO a) -> IO a
+withBytes = withPtr
+
+bytesEq :: Bytes -> Bytes -> Bool
+bytesEq b1@(Bytes m1) b2@(Bytes m2)
+    | l1 /= l2  = False
+    | otherwise = unsafeDoIO $ IO $ \s -> loop 0# s
+  where
+    !l1@(I# len) = bytesLength b1
+    !l2          = bytesLength b2
+
+    loop i s
+        | booleanPrim (i ==# len) = (# s, True #)
+        | otherwise               =
+            case readWord8Array# m1 i s of
+                (# s', e1 #) -> case readWord8Array# m2 i s' of
+                    (# s'', e2 #) ->
+                        if booleanPrim (eqWord# e1 e2)
+                            then loop (i +# 1#) s''
+                            else (# s', False #)
+
+{-
+bytesIndex :: Bytes -> Int -> Word8
+bytesIndex (Bytes m) (I# i) = unsafeDoIO $ IO $ \s ->
+    case readWord8Array# m i s of
+        (# s', e #) -> (# s', W8# e #)
+{-# NOINLINE bytesIndex #-}
+-}
+
+bytesShowHex :: Bytes -> String
+bytesShowHex b = showHexadecimal (withPtr b) (bytesLength b)
+{-# NOINLINE bytesShowHex #-}
diff --git a/Data/Memory/ByteArray/MemView.hs b/Data/Memory/ByteArray/MemView.hs
new file mode 100644
index 0000000..5663d60
--- /dev/null
+++ b/Data/Memory/ByteArray/MemView.hs
@@ -0,0 +1,21 @@
+-- |
+-- Module      : Data.Memory.ByteArray.MemView
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : stable
+-- Portability : Good
+--
+module Data.Memory.ByteArray.MemView
+    ( MemView(..)
+    ) where
+
+import           Foreign.Ptr
+import           Data.Memory.ByteArray.Types
+import           Data.Memory.Internal.Imports
+
+data MemView = MemView !(Ptr Word8) !Int
+
+instance ByteArrayAccess MemView where
+    length (MemView _ l) = l
+    withByteArray (MemView p _) f = f (castPtr p)
+
diff --git a/Data/Memory/ByteArray/Methods.hs b/Data/Memory/ByteArray/Methods.hs
new file mode 100644
index 0000000..9c5ea3e
--- /dev/null
+++ b/Data/Memory/ByteArray/Methods.hs
@@ -0,0 +1,191 @@
+-- |
+-- Module      : Data.Memory.ByteArray.Methods
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : stable
+-- Portability : Good
+--
+{-# LANGUAGE BangPatterns #-}
+module Data.Memory.ByteArray.Methods
+    ( alloc
+    , allocAndFreeze
+    , empty
+    , zero
+    , copy
+    , take
+    , convert
+    , convertHex
+    , copyRet
+    , copyAndFreeze
+    , split
+    , xor
+    , eq
+    , index
+    , constEq
+    , concat
+    , toW64BE
+    , toW64LE
+    , mapAsWord64
+    , mapAsWord128
+    ) where
+
+import           Data.Memory.Internal.Compat
+import           Data.Memory.Internal.Imports hiding (empty)
+import           Data.Memory.ByteArray.Types
+import           Data.Memory.Endian
+import           Data.Memory.PtrMethods
+import           Data.Memory.ExtendedWords
+import           Data.Memory.Encoding.Base16
+import           Foreign.Storable
+import           Foreign.Ptr
+
+import           Prelude hiding (length, take, concat)
+
+alloc :: ByteArray ba => Int -> (Ptr p -> IO ()) -> IO ba
+alloc n f = snd `fmap` allocRet n f
+
+allocAndFreeze :: ByteArray a => Int -> (Ptr p -> IO ()) -> a
+allocAndFreeze sz f = unsafeDoIO (alloc sz f)
+
+empty :: ByteArray a => a
+empty = unsafeDoIO (alloc 0 $ \_ -> return ())
+
+-- | Create a xor of bytes between a and b.
+--
+-- the returns byte array is the size of the smallest input.
+xor :: (ByteArrayAccess a, ByteArrayAccess b, ByteArray c) => a -> b -> c
+xor a b =
+    allocAndFreeze n $ \pc ->
+    withByteArray a  $ \pa ->
+    withByteArray b  $ \pb ->
+        memXor pc pa pb n
+  where
+        n  = min la lb
+        la = length a
+        lb = length b
+
+index :: ByteArrayAccess a => a -> Int -> Word8
+index b i = unsafeDoIO $ withByteArray b $ \p -> peek (p `plusPtr` i)
+
+split :: ByteArray bs => Int -> bs -> (bs, bs)
+split n bs
+    | n <= 0    = (empty, bs)
+    | n >= len  = (bs, empty)
+    | otherwise = unsafeDoIO $ do
+        withByteArray bs $ \p -> do
+            b1 <- alloc n $ \r -> memCopy r p n
+            b2 <- alloc (len - n) $ \r -> memCopy r (p `plusPtr` n) (len - n)
+            return (b1, b2)
+  where len = length bs
+
+take :: ByteArray bs => Int -> bs -> bs
+take n bs =
+    allocAndFreeze m $ \d -> withByteArray bs $ \s -> memCopy d s m
+  where
+        m   = min len n
+        len = length bs
+
+concat :: ByteArray bs => [bs] -> bs
+concat []    = empty
+concat allBs = allocAndFreeze total (loop allBs)
+  where
+        total = sum $ map length allBs
+
+        loop []     _   = return ()
+        loop (b:bs) dst = do
+            let sz = length b
+            withByteArray b $ \p -> memCopy dst p sz
+            loop bs (dst `plusPtr` sz)
+
+copy :: (ByteArrayAccess bs1, ByteArray bs2) => bs1 -> (Ptr p -> IO ()) -> IO bs2
+copy bs f =
+    alloc (length bs) $ \d -> do
+        withByteArray bs $ \s -> memCopy d s (length bs)
+        f (castPtr d)
+
+copyRet :: (ByteArrayAccess bs1, ByteArray bs2) => bs1 -> (Ptr p -> IO a) -> IO (a, bs2)
+copyRet bs f =
+    allocRet (length bs) $ \d -> do
+        withByteArray bs $ \s -> memCopy d s (length bs)
+        f (castPtr d)
+
+copyAndFreeze :: (ByteArrayAccess bs1, ByteArray bs2) => bs1 -> (Ptr p -> IO ()) -> bs2
+copyAndFreeze bs f =
+    allocAndFreeze (length bs) $ \d -> do
+        withByteArray bs $ \s -> memCopy d s (length bs)
+        f (castPtr d)
+
+zero :: ByteArray ba => Int -> ba
+zero n = allocAndFreeze n $ \ptr -> memSet ptr 0 n
+
+eq :: (ByteArrayAccess bs1, ByteArrayAccess bs2) => bs1 -> bs2 -> Bool
+eq b1 b2
+    | l1 /= l2  = False
+    | otherwise = unsafeDoIO $ withByteArray b1 $ \p1 -> withByteArray b2 $ \p2 -> memEqual p1 p2 l1
+  where
+    l1 = length b1
+    l2 = length b2
+
+-- | A constant time equality test for 2 ByteArrayAccess values.
+--
+-- If values are of 2 different sizes, the function will abort early
+-- without comparing any bytes.
+--
+-- compared to == , this function will go over all the bytes
+-- present before yielding a result even when knowing the
+-- overall result early in the processing.
+constEq :: (ByteArrayAccess bs1, ByteArrayAccess bs2) => bs1 -> bs2 -> Bool
+constEq b1 b2
+    | l1 /= l2  = False
+    | otherwise = unsafeDoIO $ withByteArray b1 $ \p1 -> withByteArray b2 $ \p2 -> memConstEqual p1 p2 l1
+  where
+    l1 = length b1
+    l2 = length b2
+
+toW64BE :: ByteArrayAccess bs => bs -> Int -> BE Word64
+toW64BE bs ofs = unsafeDoIO $ withByteArray bs $ \p -> peek (p `plusPtr` ofs)
+
+toW64LE :: ByteArrayAccess bs => bs -> Int -> LE Word64
+toW64LE bs ofs = unsafeDoIO $ withByteArray bs $ \p -> peek (p `plusPtr` ofs)
+
+mapAsWord128 :: ByteArray bs => (Word128 -> Word128) -> bs -> bs
+mapAsWord128 f bs =
+    allocAndFreeze len $ \dst ->
+    withByteArray bs   $ \src ->
+        loop (len `div` 16) dst src
+  where
+        len        = length bs
+        loop :: Int -> Ptr (BE Word64) -> Ptr (BE Word64) -> IO ()
+        loop 0 _ _ = return ()
+        loop i d s = do
+            w1 <- peek s
+            w2 <- peek (s `plusPtr` 8)
+            let (Word128 r1 r2) = f (Word128 (fromBE w1) (fromBE w2))
+            poke d               (toBE r1)
+            poke (d `plusPtr` 8) (toBE r2)
+            loop (i-1) (d `plusPtr` 16) (s `plusPtr` 16)
+
+mapAsWord64 :: ByteArray bs => (Word64 -> Word64) -> bs -> bs
+mapAsWord64 f bs =
+    allocAndFreeze len $ \dst ->
+    withByteArray bs            $ \src ->
+        loop (len `div` 8) dst src
+  where
+        len        = length bs
+
+        loop :: Int -> Ptr (BE Word64) -> Ptr (BE Word64) -> IO ()
+        loop 0 _ _ = return ()
+        loop i d s = do
+            w <- peek s
+            let r = f (fromBE w)
+            poke d (toBE r)
+            loop (i-1) (d `plusPtr` 8) (s `plusPtr` 8)
+
+convert :: (ByteArrayAccess bin, ByteArray bout) => bin -> bout
+convert = flip copyAndFreeze (\_ -> return ())
+
+convertHex :: (ByteArrayAccess bin, ByteArray bout) => bin -> bout
+convertHex b =
+    allocAndFreeze (length b * 2) $ \bout ->
+    withByteArray b               $ \bin  ->
+        toHexadecimal bout bin (length b)
diff --git a/Data/Memory/ByteArray/ScrubbedBytes.hs b/Data/Memory/ByteArray/ScrubbedBytes.hs
new file mode 100644
index 0000000..7840122
--- /dev/null
+++ b/Data/Memory/ByteArray/ScrubbedBytes.hs
@@ -0,0 +1,103 @@
+-- |
+-- Module      : Data.Memory.ByteArray.ScrubbedBytes
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : Stable
+-- Portability : GHC
+--
+{-# LANGUAGE BangPatterns #-}
+{-# LANGUAGE MagicHash #-}
+{-# LANGUAGE UnboxedTuples #-}
+{-# LANGUAGE CPP #-}
+module Data.Memory.ByteArray.ScrubbedBytes
+    ( ScrubbedBytes
+    ) where
+
+import           GHC.Types
+import           GHC.Prim
+import           GHC.Ptr
+import           Data.Memory.Internal.CompatPrim
+import           Data.Memory.Internal.Compat     (unsafeDoIO)
+import           Data.Memory.PtrMethods          (memConstEqual)
+import           Data.Memory.ByteArray.Types
+
+-- | ScrubbedBytes is a memory chunk which have the properties of:
+--
+-- * Being scrubbed after its goes out of scope.
+--
+-- * A Show instance that doesn't actually show any content
+--
+-- * A Eq instance that is constant time
+--
+data ScrubbedBytes = ScrubbedBytes (MutableByteArray# RealWorld)
+
+instance Show ScrubbedBytes where
+    show _ = "<scrubbed-bytes>"
+
+instance Eq ScrubbedBytes where
+    (==) = scrubbedBytesEq
+
+instance ByteArrayAccess ScrubbedBytes where
+    length        = sizeofScrubbedBytes
+    withByteArray = withPtr
+
+instance ByteArray ScrubbedBytes where
+    allocRet = scrubbedBytesAllocRet
+
+newScrubbedBytes :: Int -> IO ScrubbedBytes
+newScrubbedBytes (I# sz)
+    | booleanPrim (sz <=# 0#) = error "negative or null size for scrubbed array" -- TODO raise a proper exception
+    | otherwise               = IO $ \s ->
+        case newAlignedPinnedByteArray# sz 8# s of
+            (# s1, mbarr #) ->
+                let !scrubber = getScrubber
+                    !mba      = ScrubbedBytes mbarr
+                 in case mkWeak# mbarr () (scrubber (byteArrayContents# (unsafeCoerce# mbarr)) >> touchScrubbedBytes mba) s1 of
+                    (# s2, _ #) -> (# s2, mba #)
+  where
+        getScrubber :: Addr# -> IO ()
+        getScrubber = eitherDivideBy8# sz scrubber64 scrubber8
+
+        scrubber64 :: Int# -> Addr# -> IO ()
+        scrubber64 sz64 addr = IO $ \s -> (# loop sz64 addr s, () #)
+          where loop :: Int# -> Addr# -> State# RealWorld -> State# RealWorld
+                loop n a s
+                    | booleanPrim (n ==# 0#) = s
+                    | otherwise              =
+                        case writeWord64OffAddr# a 0# 0## s of
+                            s' -> loop (n -# 1#) (plusAddr# a 8#) s'
+
+        scrubber8 :: Int# -> Addr# -> IO ()
+        scrubber8 sz8 addr = IO $ \s -> (# loop sz8 addr s, () #)
+          where loop :: Int# -> Addr# -> State# RealWorld -> State# RealWorld
+                loop n a s
+                    | booleanPrim (n ==# 0#) = s
+                    | otherwise              =
+                        case writeWord8OffAddr# a 0# 0## s of
+                            s' -> loop (n -# 1#) (plusAddr# a 1#) s'
+
+scrubbedBytesAllocRet :: Int -> (Ptr p -> IO a) -> IO (a, ScrubbedBytes)
+scrubbedBytesAllocRet sz f = do
+    ba <- newScrubbedBytes sz
+    r  <- withPtr ba f
+    return (r, ba)
+
+sizeofScrubbedBytes :: ScrubbedBytes -> Int
+sizeofScrubbedBytes (ScrubbedBytes mba) = I# (sizeofMutableByteArray# mba)
+
+withPtr :: ScrubbedBytes -> (Ptr p -> IO a) -> IO a
+withPtr b@(ScrubbedBytes mba) f = do
+    a <- f (Ptr (byteArrayContents# (unsafeCoerce# mba)))
+    touchScrubbedBytes b
+    return a
+
+touchScrubbedBytes :: ScrubbedBytes -> IO ()
+touchScrubbedBytes (ScrubbedBytes mba) = IO $ \s -> case touch# mba s of s' -> (# s', () #)
+
+scrubbedBytesEq :: ScrubbedBytes -> ScrubbedBytes -> Bool
+scrubbedBytesEq a b
+    | l1 /= l2  = False
+    | otherwise = unsafeDoIO $ withPtr a $ \p1 -> withPtr b $ \p2 -> memConstEqual p1 p2 l1
+  where
+        l1 = sizeofScrubbedBytes a
+        l2 = sizeofScrubbedBytes b
diff --git a/Data/Memory/ByteArray/Types.hs b/Data/Memory/ByteArray/Types.hs
new file mode 100644
index 0000000..143e76f
--- /dev/null
+++ b/Data/Memory/ByteArray/Types.hs
@@ -0,0 +1,41 @@
+-- |
+-- Module      : Data.Memory.ByteArray.Types
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : stable
+-- Portability : Good
+--
+{-# LANGUAGE CPP #-}
+module Data.Memory.ByteArray.Types
+    ( ByteArrayAccess(..)
+    , ByteArray(..)
+    ) where
+
+import           Foreign.Ptr
+
+#ifdef WITH_BYTESTRING_SUPPORT
+import qualified Data.ByteString as B (length)
+import qualified Data.ByteString.Internal as B
+import           Foreign.ForeignPtr (withForeignPtr)
+#endif
+
+class ByteArrayAccess ba where
+    length        :: ba -> Int
+    withByteArray :: ba -> (Ptr p -> IO a) -> IO a
+
+class ByteArrayAccess ba => ByteArray ba where
+    allocRet  :: Int -> (Ptr p -> IO a) -> IO (a, ba)
+
+#ifdef WITH_BYTESTRING_SUPPORT
+instance ByteArrayAccess B.ByteString where
+    length = B.length
+    withByteArray b f = withForeignPtr fptr $ \ptr -> f (ptr `plusPtr` off)
+      where (fptr, off, _) = B.toForeignPtr b
+
+instance ByteArray B.ByteString where
+    allocRet sz f = do
+        fptr <- B.mallocByteString sz
+        r    <- withForeignPtr fptr (f . castPtr)
+        return (r, B.PS fptr 0 sz)
+#endif
+
diff --git a/Data/Memory/Encoding/Base16.hs b/Data/Memory/Encoding/Base16.hs
new file mode 100644
index 0000000..d7a3e9b
--- /dev/null
+++ b/Data/Memory/Encoding/Base16.hs
@@ -0,0 +1,99 @@
+-- |
+-- Module      : Data.Memory.Encoding.Base16
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : experimental
+-- Portability : unknown
+--
+-- Hexadecimal escaper
+--
+{-# LANGUAGE MagicHash #-}
+{-# LANGUAGE UnboxedTuples #-}
+{-# LANGUAGE OverloadedStrings #-}
+{-# LANGUAGE BangPatterns #-}
+{-# LANGUAGE Rank2Types #-}
+module Data.Memory.Encoding.Base16
+    ( showHexadecimal
+    , toHexadecimal
+    ) where
+
+import           Data.Memory.Internal.Compat
+import           Data.Word
+import           GHC.Prim
+import           GHC.Types
+import           GHC.Word
+import           Control.Monad
+import           Foreign.Storable
+import           Foreign.Ptr (Ptr)
+
+showHexadecimal :: (forall a . (Ptr Word8 -> IO a) -> IO a)
+                -> Int
+                -> String
+showHexadecimal withPtr = doChunks 0
+  where
+        doChunks ofs len
+            | len < 4   = doUnique ofs len
+            | otherwise = do
+                let !(W8# a, W8# b, W8# c, W8# d) = unsafeDoIO $ withPtr (read4 ofs)
+                    !(# w1, w2 #) = convertByte a
+                    !(# w3, w4 #) = convertByte b
+                    !(# w5, w6 #) = convertByte c
+                    !(# w7, w8 #) = convertByte d
+                 in wToChar w1 : wToChar w2 : wToChar w3 : wToChar w4
+                  : wToChar w5 : wToChar w6 : wToChar w7 : wToChar w8
+                  : doChunks (ofs + 4) (len - 4)
+
+        doUnique ofs len
+            | len == 0  = []
+            | otherwise =
+                let !(W8# b)      = unsafeDoIO $ withPtr (byteIndex ofs)
+                    !(# w1, w2 #) = convertByte b
+                 in wToChar w1 : wToChar w2 : doUnique (ofs + 1) (len - 1)
+
+        read4 :: Int -> Ptr Word8 -> IO (Word8, Word8, Word8, Word8)
+        read4 ofs p =
+            liftM4 (,,,) (byteIndex ofs     p) (byteIndex (ofs+1) p)
+                         (byteIndex (ofs+2) p) (byteIndex (ofs+3) p)
+
+        wToChar :: Word# -> Char
+        wToChar w = toEnum (I# (word2Int# w))
+
+        byteIndex :: Int -> Ptr Word8 -> IO Word8
+        byteIndex i p = peekByteOff p i
+
+toHexadecimal :: Ptr Word8 -> Ptr Word8 -> Int -> IO ()
+toHexadecimal bout bin n = loop 0
+  where loop i
+            | i == n  = return ()
+            | otherwise = do
+                (W8# w) <- peekByteOff bin i
+                let !(# w1, w2 #) = convertByte w
+                pokeByteOff bout (i * 2)     (W8# w1)
+                pokeByteOff bout (i * 2 + 1) (W8# w2)
+                loop (i+1)
+
+convertByte :: Word# -> (# Word#, Word# #)
+convertByte b = (# r tableHi b, r tableLo b #)
+  where
+        r :: Addr# -> Word# -> Word#
+        r table index = indexWord8OffAddr# table (word2Int# index)
+
+        !tableLo =
+            "0123456789abcdef0123456789abcdef\
+            \0123456789abcdef0123456789abcdef\
+            \0123456789abcdef0123456789abcdef\
+            \0123456789abcdef0123456789abcdef\
+            \0123456789abcdef0123456789abcdef\
+            \0123456789abcdef0123456789abcdef\
+            \0123456789abcdef0123456789abcdef\
+            \0123456789abcdef0123456789abcdef"#
+        !tableHi =
+            "00000000000000001111111111111111\
+            \22222222222222223333333333333333\
+            \44444444444444445555555555555555\
+            \66666666666666667777777777777777\
+            \88888888888888889999999999999999\
+            \aaaaaaaaaaaaaaaabbbbbbbbbbbbbbbb\
+            \ccccccccccccccccdddddddddddddddd\
+            \eeeeeeeeeeeeeeeeffffffffffffffff"#
+{-# INLINE convertByte #-}
diff --git a/Data/Memory/Endian.hs b/Data/Memory/Endian.hs
new file mode 100644
index 0000000..0735194
--- /dev/null
+++ b/Data/Memory/Endian.hs
@@ -0,0 +1,114 @@
+-- |
+-- Module      : Data.Memory.Endian
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : stable
+-- Portability : good
+--
+{-# LANGUAGE CPP #-}
+{-# LANGUAGE GeneralizedNewtypeDeriving #-}
+module Data.Memory.Endian
+    ( Endianness(..)
+    , getSystemEndianness
+    , BE(..), LE(..)
+    , fromBE, toBE
+    , fromLE, toLE
+    ) where
+
+import Data.Word (Word16, Word32, Word64)
+import Foreign.Storable
+#if !defined(ARCH_IS_LITTLE_ENDIAN) && !defined(ARCH_IS_BIG_ENDIAN)
+import Data.Memory.Internal.Compat (unsafeDoIO)
+#endif
+
+import Data.Memory.Internal.Compat (byteSwap64, byteSwap32, byteSwap16)
+
+-- | represent the CPU endianness
+--
+-- Big endian system stores bytes with the MSB as the first byte.
+-- Little endian system stores bytes with the LSB as the first byte.
+--
+-- middle endian is purposely avoided.
+data Endianness = LittleEndian
+                | BigEndian
+                deriving (Show,Eq)
+
+-- | Return the system endianness
+getSystemEndianness :: Endianness
+#ifdef ARCH_IS_LITTLE_ENDIAN
+getSystemEndianness = LittleEndian
+#elif ARCH_IS_BIG_ENDIAN
+getSystemEndianness = BigEndian
+#else
+getSystemEndianness
+    | isLittleEndian = LittleEndian
+    | isBigEndian    = BigEndian
+    | otherwise      = error "cannot determine endianness"
+  where
+        isLittleEndian = endianCheck == 2
+        isBigEndian    = endianCheck == 1
+        endianCheck    = unsafeDoIO $ alloca $ \p -> do
+                            poke p (0x01000002 :: Word32)
+                            peek (castPtr p :: Ptr Word8)
+#endif
+
+-- | Little Endian value
+newtype LE a = LE { unLE :: a }
+    deriving (Show,Eq,Storable)
+
+-- | Big Endian value
+newtype BE a = BE { unBE :: a }
+    deriving (Show,Eq,Storable)
+
+-- | Convert a value in cpu endianess to big endian
+toBE :: ByteSwap a => a -> BE a
+#ifdef ARCH_IS_LITTLE_ENDIAN
+toBE = BE . byteSwap
+#elif ARCH_IS_BIG_ENDIAN
+toBE = BE
+#else
+toBE = BE . (if getSystemEndianness == LittleEndian then byteSwap else id)
+#endif
+{-# INLINE toBE #-}
+
+-- | Convert from a big endian value to the cpu endianness
+fromBE :: ByteSwap a => BE a -> a
+#ifdef ARCH_IS_LITTLE_ENDIAN
+fromBE (BE a) = byteSwap a
+#elif ARCH_IS_BIG_ENDIAN
+fromBE (BE a) = a
+#else
+fromBE (BE a) = if getSystemEndianness == LittleEndian then byteSwap a else a
+#endif
+{-# INLINE fromBE #-}
+
+-- | Convert a value in cpu endianess to little endian
+toLE :: ByteSwap a => a -> LE a
+#ifdef ARCH_IS_LITTLE_ENDIAN
+toLE = LE
+#elif ARCH_IS_BIG_ENDIAN
+toLE = LE . byteSwap
+#else
+toLE = LE . (if getSystemEndianness == LittleEndian then id else byteSwap)
+#endif
+{-# INLINE toLE #-}
+
+-- | Convert from a little endian value to the cpu endianness
+fromLE :: ByteSwap a => LE a -> a
+#ifdef ARCH_IS_LITTLE_ENDIAN
+fromLE (LE a) = a
+#elif ARCH_IS_BIG_ENDIAN
+fromLE (LE a) = byteSwap a
+#else
+fromLE (LE a) = if getSystemEndianness == LittleEndian then a else byteSwap a
+#endif
+{-# INLINE fromLE #-}
+
+class Storable a => ByteSwap a where
+    byteSwap :: a -> a
+instance ByteSwap Word16 where
+    byteSwap = byteSwap16
+instance ByteSwap Word32 where
+    byteSwap = byteSwap32
+instance ByteSwap Word64 where
+    byteSwap = byteSwap64
diff --git a/Data/Memory/ExtendedWords.hs b/Data/Memory/ExtendedWords.hs
new file mode 100644
index 0000000..6e2052b
--- /dev/null
+++ b/Data/Memory/ExtendedWords.hs
@@ -0,0 +1,16 @@
+-- |
+-- Module      : Data.Memory.ExtendedWords
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : experimental
+-- Portability : unknown
+--
+-- Extra Word size
+--
+module Data.Memory.ExtendedWords
+    ( Word128(..)
+    ) where
+
+import Data.Word (Word64)
+
+data Word128 = Word128 !Word64 !Word64 deriving (Show, Eq)
diff --git a/Data/Memory/Internal/Compat.hs b/Data/Memory/Internal/Compat.hs
new file mode 100644
index 0000000..2b94112
--- /dev/null
+++ b/Data/Memory/Internal/Compat.hs
@@ -0,0 +1,65 @@
+-- |
+-- Module      : Data.Memory.Internal.Compat
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : stable
+-- Portability : Good
+--
+-- This module try to keep all the difference between versions of base
+-- or other needed packages, so that modules don't need to use CPP
+--
+{-# LANGUAGE CPP #-}
+module Data.Memory.Internal.Compat
+    ( unsafeDoIO
+    , popCount
+    , byteSwap64
+    , byteSwap32
+    , byteSwap16
+    ) where
+
+import System.IO.Unsafe
+import Data.Word
+import Data.Bits
+
+-- | perform io for hashes that do allocation and ffi.
+-- unsafeDupablePerformIO is used when possible as the
+-- computation is pure and the output is directly linked
+-- to the input. we also do not modify anything after it has
+-- been returned to the user.
+unsafeDoIO :: IO a -> a
+#if __GLASGOW_HASKELL__ > 704
+unsafeDoIO = unsafeDupablePerformIO
+#else
+unsafeDoIO = unsafePerformIO
+#endif
+
+#if !(MIN_VERSION_base(4,5,0))
+popCount :: Word64 -> Int
+popCount n = loop 0 n
+  where loop c 0 = c
+        loop c i = loop (c + if testBit c 0 then 1 else 0) (i `shiftR` 1)
+#endif
+
+#if !(MIN_VERSION_base(4,7,0))
+byteSwap64 :: Word64 -> Word64
+byteSwap64 w =
+        (w `shiftR` 56)                  .|. (w `shiftL` 56)
+    .|. ((w `shiftR` 40) .&. 0xff00)     .|. ((w .&. 0xff00) `shiftL` 40)
+    .|. ((w `shiftR` 24) .&. 0xff0000)   .|. ((w .&. 0xff0000) `shiftL` 24)
+    .|. ((w `shiftR` 8)  .&. 0xff000000) .|. ((w .&. 0xff000000) `shiftL` 8)
+#endif
+
+#if !(MIN_VERSION_base(4,7,0))
+byteSwap32 :: Word32 -> Word32
+byteSwap32 w =
+        (w `shiftR` 24)
+    .|. (w `shiftL` 24)
+    .|. ((w `shiftR` 8) .&. 0xff00)
+    .|. ((w .&. 0xff00) `shiftL` 8)
+#endif
+
+#if !(MIN_VERSION_base(4,7,0))
+byteSwap16 :: Word16 -> Word16
+byteSwap16 w =
+    (w `shiftR` 8) .|. (w `shiftL` 8)
+#endif
diff --git a/Data/Memory/Internal/CompatPrim.hs b/Data/Memory/Internal/CompatPrim.hs
new file mode 100644
index 0000000..1b7bb15
--- /dev/null
+++ b/Data/Memory/Internal/CompatPrim.hs
@@ -0,0 +1,84 @@
+-- |
+-- Module      : Data.Memory.Internal.CompatPrim
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : stable
+-- Portability : Compat
+--
+-- This module try to keep all the difference between versions of ghc primitive
+-- or other needed packages, so that modules don't need to use CPP.
+--
+-- Note that MagicHash and CPP conflicts in places, making it "more interesting"
+-- to write compat code for primitives
+--
+{-# LANGUAGE CPP #-}
+{-# LANGUAGE BangPatterns #-}
+{-# LANGUAGE MagicHash #-}
+{-# LANGUAGE UnboxedTuples #-}
+module Data.Memory.Internal.CompatPrim
+    ( be32Prim
+    , le32Prim
+    , byteswap32Prim
+    , booleanPrim
+    , eitherDivideBy8#
+    ) where
+
+import GHC.Prim
+
+-- | byteswap Word# to or from Big Endian
+--
+-- on a big endian machine, this function is a nop.
+be32Prim :: Word# -> Word#
+#ifdef ARCH_IS_LITTLE_ENDIAN
+be32Prim = byteswap32Prim
+#else
+be32Prim w = w
+#endif
+
+-- | byteswap Word# to or from Little Endian
+--
+-- on a little endian machine, this function is a nop.
+le32Prim :: Word# -> Word#
+#ifdef ARCH_IS_LITTLE_ENDIAN
+le32Prim w = w
+#else
+le32Prim = byteswap32Prim
+#endif
+
+byteswap32Prim :: Word# -> Word#
+#if __GLASGOW_HASKELL__ >= 708
+byteswap32Prim w = byteSwap32# w
+#else
+byteswap32Prim w =
+    let !a =       uncheckedShiftL# w 24#
+        !b = and# (uncheckedShiftL# w 8#) 0x00ff0000##
+        !c = and# (uncheckedShiftRL# w 8#) 0x0000ff00##
+        !d = and# (uncheckedShiftRL# w 24#) 0x000000ff##
+     in or# a (or# b (or# c d))
+#endif
+
+#if __GLASGOW_HASKELL__ >= 708
+booleanPrim :: Int# -> Bool
+booleanPrim v = tagToEnum# v
+#else
+booleanPrim :: Bool -> Bool
+booleanPrim b = b
+#endif
+
+-- | Apply or or another function if 8 divides the number of bytes
+eitherDivideBy8# :: Int#        -- ^ number of bytes
+                 -> (Int# -> a) -- ^ if it divided by 8, the argument is the number of 8 bytes words
+                 -> (Int# -> a) -- ^ if it doesn't, just the number of bytes
+                 -> a
+#if __GLASGOW_HASKELL__ >= 740
+eitherDivideBy8# v f8 f1 =
+    let !(# q, r #) = quotRemInt v 8#
+     in if booleanPrim (r ==# 0)
+            then f8 q
+            else f1 v
+#else
+eitherDivideBy8# v f8 f1 =
+    if booleanPrim ((remInt# v 8#) ==# 0#)
+        then f8 (quotInt# v 8#)
+        else f1 v
+#endif
diff --git a/Data/Memory/Internal/Imports.hs b/Data/Memory/Internal/Imports.hs
new file mode 100644
index 0000000..6a4d830
--- /dev/null
+++ b/Data/Memory/Internal/Imports.hs
@@ -0,0 +1,15 @@
+-- |
+-- Module      : Data.Memory.Internal.Imports
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : experimental
+-- Portability : unknown
+--
+module Data.Memory.Internal.Imports
+    ( module X
+    ) where
+
+import Data.Word           as X
+import Control.Applicative as X
+import Control.Monad       as X (forM, forM_, void)
+import Control.Arrow       as X (first, second)
diff --git a/Data/Memory/PtrMethods.hs b/Data/Memory/PtrMethods.hs
new file mode 100644
index 0000000..eb940c3
--- /dev/null
+++ b/Data/Memory/PtrMethods.hs
@@ -0,0 +1,110 @@
+-- |
+-- Module      : Data.Memory.PtrMethods
+-- License     : BSD-style
+-- Maintainer  : Vincent Hanquez <vincent@snarc.org>
+-- Stability   : experimental
+-- Portability : unknown
+--
+-- methods to manipulate raw memory representation
+--
+{-# LANGUAGE BangPatterns #-}
+{-# LANGUAGE MagicHash #-}
+{-# LANGUAGE UnboxedTuples #-}
+{-# LANGUAGE ForeignFunctionInterface #-}
+module Data.Memory.PtrMethods
+    ( memCreateTemporary
+    , memXor
+    , memXorWith
+    , memCopy
+    , memSet
+    , memEqual
+    , memConstEqual
+    , memCompare
+    ) where
+
+import           Data.Memory.Internal.Imports
+import           Foreign.Ptr              (Ptr, plusPtr)
+import           Foreign.Storable         (peek, poke, pokeByteOff, peekByteOff)
+import           Foreign.C.Types
+import           Foreign.Marshal.Alloc    (allocaBytesAligned)
+import           Data.Bits                (xor)
+
+-- | Create a new temporary buffer
+memCreateTemporary :: Int -> (Ptr Word8 -> IO a) -> IO a
+memCreateTemporary size f = allocaBytesAligned size 8 f
+
+-- | xor bytes from source1 and source2 to destination
+-- 
+-- d = s1 xor s2
+--
+-- s1, nor s2 are modified unless d point to s1 or s2
+memXor :: Ptr Word8 -> Ptr Word8 -> Ptr Word8 -> Int -> IO ()
+memXor _ _  _  0 = return ()
+memXor d s1 s2 n = do
+    (xor <$> peek s1 <*> peek s2) >>= poke d
+    memXor (d `plusPtr` 1) (s1 `plusPtr` 1) (s2 `plusPtr` 1) (n-1)
+
+-- | xor bytes from source with a specific value to destination
+--
+-- d = replicate (sizeof s) v `xor` s
+memXorWith :: Ptr Word8 -> Word8 -> Ptr Word8 -> Int -> IO ()
+memXorWith d v s n = loop 0
+  where
+    loop i
+        | i == n    = return ()
+        | otherwise = do
+            (xor v <$> peekByteOff s i) >>= pokeByteOff d i
+            loop (i+1)
+
+-- | Copy a set number of bytes from @src to @dst
+memCopy :: Ptr Word8 -> Ptr Word8 -> Int -> IO ()
+memCopy dst src n = c_memcpy dst src (fromIntegral n)
+
+-- | Set @n number of bytes to the same value @v
+memSet :: Ptr Word8 -> Word8 -> Int -> IO ()
+memSet start v n = c_memset start (fromIntegral v) (fromIntegral n) >>= \_ -> return ()
+
+memEqual :: Ptr Word8 -> Ptr Word8 -> Int -> IO Bool
+memEqual p1 p2 n = loop 0
+  where
+    loop i
+        | i == n    = return True
+        | otherwise = do
+            e <- (==) <$> peekByteOff p1 i <*> (peekByteOff p2 i :: IO Word8)
+            if e then loop (i+1) else return False
+
+memCompare :: Ptr Word8 -> Ptr Word8 -> Int -> IO Ordering
+memCompare p1 p2 n = loop 0
+  where
+    loop i
+        | i == n    = return EQ
+        | otherwise = do
+            e <- compare <$> peekByteOff p1 i <*> (peekByteOff p2 i :: IO Word8)
+            if e == EQ then loop (i+1) else return e
+
+-- | A constant time equality test for 2 Memory buffers
+--
+-- compared to normal equality function, this function will go
+-- over all the bytes present before yielding a result even when
+-- knowing the overall result early in the processing.
+memConstEqual :: Ptr Word8 -> Ptr Word8 -> Int -> IO Bool
+memConstEqual p1 p2 n = loop 0 True
+  where
+    loop i !ret
+        | i == n    = return ret
+        | otherwise = do
+            e <- (==) <$> peek p1 <*> peek p2
+            loop (i+1) (ret &&! e)
+
+    -- Bool == Bool
+    (&&!) :: Bool -> Bool -> Bool
+    True  &&! True  = True
+    True  &&! False = False
+    False &&! True  = False
+    False &&! False = False
+
+foreign import ccall unsafe "memset"
+    c_memset :: Ptr Word8 -> Word8 -> CSize -> IO ()
+
+foreign import ccall unsafe "memcpy"
+    c_memcpy :: Ptr Word8 -> Ptr Word8 -> CSize -> IO ()
diff --git a/cryptonite.cabal b/cryptonite.cabal
index 002445d..3b5ac23 100644
--- a/cryptonite.cabal
+++ b/cryptonite.cabal
@@ -35,6 +35,11 @@ Flag support_pclmuldq
   Default:           False
   Manual:            True
 
+Flag builtin_memory
+  Description:       Build with a local snapshot of the memory package
+  Default:           True
+  Manual:            True
+
 Library
   Exposed-modules:   Crypto.Cipher.AES
                      Crypto.Cipher.Blowfish
@@ -164,6 +169,24 @@ Library
                    , cbits/cryptonite_scrypt.c
   include-dirs:  cbits cbits/ed25519
 
+  if flag(builtin_memory)
+    Exposed-modules:   Data.Memory.ByteArray
+                       Data.Memory.Endian
+                       Data.Memory.PtrMethods
+                       Data.Memory.ExtendedWords
+                       Data.Memory.Encoding.Base16
+    Other-modules:     Data.Memory.Internal.Compat
+                       Data.Memory.Internal.CompatPrim
+                       Data.Memory.Internal.Imports
+                       Data.Memory.ByteArray.Types
+                       Data.Memory.ByteArray.Bytes
+                       Data.Memory.ByteArray.ScrubbedBytes
+                       Data.Memory.ByteArray.Methods
+                       Data.Memory.ByteArray.MemView
+    CPP-options:       -DWITH_BYTESTRING_SUPPORT
+  else
+    build-depends:     memory
+
   -- FIXME armel or mispel is also little endian.
   -- might be a good idea to also add a runtime autodetect mode.
   -- ARCH_ENDIAN_UNKNOWN