There's not really currently an easy way to do this. As mentioned already, in many browsers you can output to Canvas and then perform pixel-level transformations, but this transition from SVG to Canvas is destructive, as you have gone from a retained-mode context to an immediate-mode context, and you therefore lose all the niceties of the retained mode API, such as being able to register event listeners on individual shapes, or have a high-level API for transforming individual shapes (or groups of shapes).
However, if you don't need the element-level event handling, you might look into the dojox.gfx library, which provides a high-level, retained-mode, SVG-inspired API for drawing shapes, but also has a Canvas backend (VML and Silverlight too). I believe it is not possible to register event listeners on individual shapes when using the Canvas output, but you would be able to register an event handler on the root canvas element. It might be possible to then apply transformations with Pixtastic, but you might need to hack into the dojox.gfx Canvas render code a bit.
You might also look into SVG filters, which is as close to native support for pixel-level, raster graphics-style manipulation as it gets with SVG.
I also believe they're currently trying to combine the two specifications somewhat to make such work possible: http://lists.w3.org/Archives/Public/public-canvas-api/2011AprJun/0117.html