Questions about RVMLearner

I have a multi-class classification task in which I'd like to determine the probability of each of my N points belonging to each of C classes.  I'd like to use RVMLearner for this, but I am having trouble finding enough information about how it works in RM to properly frame the process.  In particular, how can I save the actual output, rather than just simply observing the accuracy and other stats?  Thanks!