ichark Also, I don't want to create a matrix of images with many small textures
That might be your best solution. Although it sounds like it should be expensive, filling a space with tiles, each one of which is an Image, can be very efficient. The Image class is very lightweight, containing just the vertices, and the transform so the Matrix. The main cost of it is the texture, which you can optimise by using a smaller texture, split into tiles, to fill the space.
Once it's drawn the cost of transforming the individual ties can be further optimised by them being cached in a MeshBatch. You don't need to do anything special to benefit from this, Starling can do it automatically.