TensorFlow 2.3: load model from ModelCheckPoint callback with both custom layers and model

Question

I have wrote a custom code to build a UNet architecture. To do so I have firstly subclassed the tf.keras.layers.Layer object to define an encoder convolutional block composed by