CNN Backprogation of Convolutional Layer

I build myself a CNN with just conv layers(so far).

I used the same Algorithmen from ANNs. Because the Conv layer isnt fully connected and has shared weights im not sure how to calculate this. I tried to avarage the Δweight and use the avarage to backprop. After few testings I dont think that this works and it makes the iteration very slow(1 Iteration per s instead of 80). How do I properly backprop Conv layers? And it would also be nice if anyone had also information or a link for backprop of other layers (Pooling,...)

Thank you for your time. (Sorry for the bad english).