Several variations of these pipelines would be supported!
If we want visual output I'd introduce a graphics card with a "layout coprocessor" to traverse (binary) trees negotiating position information & a "compositor coprocessor" which combines an arbitrary number of "sprites" with a couple texture fetches or multiply-sums per clockcycle. That compositor could also do offscreen renders for optimization, hittesting, etc.
Text layout would be the main bottleneck to handle a myriad of alphabets!