Optimize Gaussian filters

Gaussian filters can be run as two separate 1-by-n and n-by-1 filters instead of a single n-by-n filters. That improves times from O(n^2) to O(n) [e.g. 3x3 Gaussian filter uses 9 operations now but could use only 6, 5x5 uses 25 instead of 10]. Not applicable to other filters.
