For a current mirror to work, the hfe of the transistors must be very tightly matched, and track over temperature and aging. This is relatively easy when all Q's are on a single, very small silicon die, but much more difficult with discrete components. There are more complex current mirror circuits that work to offset or cancel out some of these effects, but the net result still is far from what goes on inside ICs. Why do you want to take this approach? What advantage do you think it might have over conventional circuits?
ak