I think this is the same trick I eventually (too late) found to make my model tractable assuming a bivariate Normal. It certainly hadn't been my first instinct to consider rate of change with respect to correlation.