Dynamically stripping XM<L

I need to be able to dynamically strip an XML message or chunk to reduce it to only contain certain peiece of information. Now I know I can use XSL, but then I have to create a stylesheet and that isn't very dynamic. I can also use DOM API and get each piece of information through XPATH signatures. But then that doesn't allow me to strip out mutiple child elements within a parent. I was leaning more towards using SAX API to parse through the entire Document and based on dynamic requirements build the output accordingly.

I am interested in the most efficient algorithm. Any suggestions??