def custom_single_cnn(size, conv_layers=(8, 16, 32, 64), dense_layers=(512, 256), conv_dropout=0.2,
                     dense_dropout=0.2):
    model = Sequential()
    model.add(
       Conv2D(conv_layers[0], kernel_size=(3, 3), padding='same', activation='relu', input_shape=(size, size, 1)))
    model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
    if conv_dropout:
       model.add(Dropout(conv_dropout))
for conv_layer_size in conv_layers[1:]:
       model.add(Conv2D(conv_layer_size, kernel_size=(3, 3), activation='relu'))
       model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
       if conv_dropout:
           model.add(Dropout(conv_dropout))
model.add(Flatten())
for dense_layer_size in dense_layers:
       model.add(Dense(dense_layer_size, activation='relu'))
       model.add(Activation('relu'))
       if dense_dropout:
           model.add(Dropout(dense_dropout))
model.add(Dense(NCATS, activation='softmax'))
    return model

<Model Declaration>
model = custom_single_cnn(size=size,
                         conv_layers=[128, 128, 256],
                         dense_layers=[2048],
                         conv_dropout=False,
                         dense_dropout=0.10 )
model.compile(optimizer=Adam(lr=0.002), loss='categorical_crossentropy',
             metrics=[categorical_crossentropy, categorical_accuracy, top_3_accuracy])
print(model.summary())