3. Can i improve overall time if i call the functions that spend too long time once every 1 msec say ?
for example i only scan discrete input every 1 msec or so.

I have found that this method, while it will in most cases speed things up add an uncertainty factor.

If you call every thing every time, the program will fail - if is to fail - whereas if you do as you suggest it may fail if some unlucky combination of interrupts and millisecond time happens.

I would stay away from such a method only because of the uncertainty it introduces.

The most costly problems are those that happen once a month, they will cost you customer confidence and a lot of scratching your head.

Another aspect is - can the "fast routines" handle not being "fast" once a millisecond.

Have fun,

Erik