We explore methods for incorporating prior knowledge about a problem at hand in Support Vector learning machines. We show that both invari(cid:173) ances under group transfonnations and prior knowledge about locality in images can be incorporated by constructing appropriate kernel functions.