It is possible to download weights for any CNN from the keras repository directly during execution of the program using :
model = VGG16(weights='imagenet', include_top=True, input_shape = (224, 224, 3))
but is it possible to download the weights trained on imagenet separately and use it to run the CNN model?