Adding a custom layer without any pretrained weights after a Conv2D layer in keras

Question

I need to add a custom layer, that is to extract only fixed number of features (say 5) from the previous Conv2D layer and fuse those 5 features using some aggregation methods (f