16 bits to 8 bits

You can just take the 8 MSBs (with rounding if you want). As long as you understand that this number is really X*256 then there is no problem. Ie. You can fit 640 into 8 bits if thats how you understand it. you can even use appropriate range definitions:

signal x : unsigned(15 downto 8);

NOTE: the numeric_std library does NOT respect bit position when doing arithmatic. You need to do this yourself by extending the arrays yourself before the functions.

The fixed point libraries do keep bit positions however.