SVM optimization with warm starts

I have a question regarding SVM software. Is there an optimization package that supports providing initial estimates for SMO?

I am doing multi-instance SVM learning, where an SVM has to be optimized very often with only slightly different labels. It would be great if I could use the solution produced for the last iteration as an initialization for the next iteration, so that the optimization doesn't need to start all over again.

I am not an expert on SMO and I'd prefer not too get to involved in the code.