Custom Deep Learning Layer

i need to implement a hybrid model of CNN and Fuzzy.

the approach is simple. by changing the last layers of the Convolutional Neural Network, exactly before the fully connected layer.

here is the structure of the proposed network

https://drive.google.com/file/d/1sq4IylLz16FBzzsXPR-Ju2F7YvQapUtD/view?usp=sharing

I have implemented CNN part but I had some troubles in the implementation of the new layer(Fuzzy Classification) in Matlab.

i had two fuzzy layers,

the first one is the FuzzyTrainingLayer which need (x,y) to train a fuzzy system.

and the second fuzzy layer is the FuzzyClasificationLayer which need the output z from FuzzyTrainingLayer(prevoiuse layer)

i have some trouble on the implementation. the first one is, how to train a fuzzy during the the main training process

the second question is how to call y(target) as well as x(from previous layer in the FuzzyTrainingLayer.

thanks in advance for your answer.(GPU_@Joss)