How to convert "dlnetwork" / "layerGraph" networks into "model function" type networks?

When I was training a custom convolutional Neural Network , the "dlnetwork" and "layerGraph" type networks had operators that I did not support, so I wanted to convert it to a "Model Function" type network. I didn't want to change the weights and biases of each layer. The data is extracted one by one (because each layer of convolution has weights and biases, manually extract them and redefine a "model function", so the manual operation is very heavy and error-prone). How to operate effectively?

Or is it best to mix "dlnetwork Object" and "Model Function" together?

---------------------------------------------------------------------------

example

Train Network Using Model Function

Train Network Using dlnetwork

But I didn’t find a mix of various types of network operations.