Abstracting markup

Hello folks, I'm toying with the idea of maximizing the flexibility of a homebrew templating utility by abstracting page markup into PHP objects. Has anybody done this successfully? I threw together a trivial demo of how this might work in practice: http://osp.claremont.edu/~dan/temp/hello.php http://osp.claremont.edu/~dan/temp/hello.phps Assuming this would only be dealing with varieties of XHTML, would the XML_Tree PEAR class be an appropriate datastructure for this situation? Comments and suggestions would be awesome! Thanks, -Dan