The HTML stuff mainly abstracts Three.js objects for faster development and better composition. Since Three.js is already an abstraction of WebGL you may have more luck with it by making a threejs plugin that "hand meshes and a camera position" to the Three.js scene/geometries.

But honestly my experience on this subject is that it would be much more manageable and productive to make the game directly with A-Frame or at least with threejs. If you need a particuliar behaviour you can go down from A-Frame to threejs and from threejs to WebGL and build the relevant components/plugins.