diff --git a/docs/architecture/assets.rst b/docs/architecture/assets.rst
new file mode 100644
index 0000000..6832ec9
--- /dev/null
+++ b/docs/architecture/assets.rst
@@ -0,0 +1,60 @@
+Asset Management
+===================================================================================================
+Layout
+---------------------------------------------------------------------------------------------------
+{version}              | 4 bytes, ie. uint32_t
+{general metadata}     | sizeof(AssetMetadata)
+{specialized metadata} | sizeof(XXXAssetMetadata), eg. TextureAssetMetadata
+{blob_0...n metadata}  | n * sizeof(BlobMetadata)
+{blob_0...n data}      | variable size based on actual data
+{end marker}           | 8 byte, ie size_t for marking the END
+
+Sections
+---------------------------------------------------------------------------------------------------
+version              -> The version of the asset for forward compatibility
+general metadata     -> Common asset metadata such as file-path, asset-type, creator, etc.
+specialized metadata -> Metadata specific to the asset, eg. texture dimensions for Textures
+blob_0...n metadata  -> Metadata specifying how the actual data is packed, required for unpacking.
+blob_0...n data      -> The actual data, packed and compressed to be reacdy for direct engine consumption.
+
+Loading
+---------------------------------------------------------------------------------------------------
+Each `Loader` has ONE OR MORE supported input file types (detected via the file extension): eg. StbLoader -> Can read in .jpg, .png, .bmp, etc.... files
+
+Each `Loader` has ONLY ONE supported output asset type:
+eg. StbLoader -> outputs TextureAsset
+
+Multiple `Loader`s MAY have as output the same asset type:
+eg. StbLoader -> outputs TextureAsset
+eg. SomeOtherImgLoader -> outputs TextureAsset
+
+Multiple `Loader`s SHOULD NOT have as input same extension types
+eg. .jpg, .png -> if supported, should only be supported by 1 `Loader` class
+
+Each `Loader` SHOULD read and turn the data from a file (eg. .png for textures) into something
+understandable by a `Packer` (not the engine itself).
+
+A `Loader` SHOULD NOT be responsible for packing the parsed file data into asset data,
+as that implies direct understanding of the layout required by the engine. 
+
+And if that layout changes, ALL `Loader`s should change accordingly; 
+which makes a class that's responsible for reading files dependant on the engine's (potentially frequent) internal changes.
+The logic is to reduce many-to-one dependency into a one-to-one dependency by redirecting the packing process to `Packer` classes
+
+Packing
+---------------------------------------------------------------------------------------------------
+Each `Packer` is responsible for packing ONLY ONE asset type:
+eg. TexturePacker for packing texture assets from parsed image files.
+eg. ModelPacker for packing model assets from parsed model files.
+
+Each `Packer` will output ONE OR MORE blobs of data, 
+and for EACH blob of data, it'll write a BlobMetadata, AFTER the specialized metadata (eg. TextureAssetMetadata)
+
+A `Packer` will make use of the `Compressor` classes to compress the data,
+and lay it out in a way that is suitable for the engine's consumption.
+
+Unpacking
+---------------------------------------------------------------------------------------------------
+A `Parser` is responsible for parsing ONLY ONE asset type:
+eg. TextureParser for parsing texture assets for direct engine consumption.
+eg. ModelParser  for parsing model assets for direct engine consumption.