Iterations vs. Kernel Size in Morphological Operations (OpenCV)

I've been using morph opening in OpenCV to reduce noise outside of my ROI in images via opencv, and until now, whenever I need a higher degree of noise reduction I just randomly increase kernel size or increase the number of iterations until I'm happy. Is there a significant difference in results depending on which you increase / how would you decide which to change in a given situation? I'm trying to come up with a better approach to which parameter I change (by how much) other than guess-and-check.