I agree regarding the use of eviews. People have made code available for both multivariate garch and Marlon switching can also be implemented. You would likely want to construct the model using their log likelihood object, which is well documented in the user manuals. Good luck.
I agree with David, that Oxmetrics 7 has some very good MS and Garch functionality. Timberlake has an econometrics summer school that, in the past, has offered training in both areas. I attended the GARCH session with Sebastian Laurent. It was excellent!! I would like to go back and tale Jurgen Doornik's MS class but it wasn't offered in 2013. The sessions with David Hendry, Jennifer Castle, Jurgen Doornik and Siem Koopman were really good too.