I think most people will say it depends. Your data structure, what you're using the calculated column or measure for, how static/dynamic they can be... several things play into it, and it's hard to know exactly what you need from what you've mentioned. For calculated column vs measures pointers, I recommend checking what some of the experts have already written. A couple of good examples are on PowerPivotPro and ExceleratorBI.
As far as calculated tables, are you talking about creating tables from scratch, like a calendar table? If you're moving to a condensed table, it sounds like you could just remove several of the rows while you're in the query editor before loading to the model. Any time you can reduce your table size in rows or columns, you'll help your performance to some degree.
