Implementation for Split-Convolution

[telefonjoker100]

I would like to have an implementation for a split convolution from pytorch down to vhdl.
This includes a part reading the values of the pytorch model, another one mapping the values for a vhdl-template and vhdl templates. The shift operation of the kernel should be solved in the Issue "Implementation for convolution". The depthwise layer should be fine, but the pointwise needs a new vhdl template