I think, you should know before, if you intend constant input or constant output data rate, or may be a variable data rate in asynchronous streams with a handshake.
Depending on this prerequisites, you should design your output interface. If a constant inpute rate of 1 byte/clock cycle would be used, you get a stream output with a length indicator anyway. You could output full bytes only (0 to 2 per cycle) and buffer the surplus bits or output a variable number of bits for each cycle.
Without a known data sink interface, the question is somewhat bloodless.