In order to make MNIPediatricAsym (or any currently unsupported volumetric template) work, we need high-quality transforms to MNI152NLin6Asym, MNI152NLin2009cAsym, or MNIInfant. I should note that the transforms we have for MNIInfant (e.g., to MNI152NLin6Asym) are low quality, which is one reason we typically run XCP-D on CIFTIs when working with infant data.
If you’re interested in helping, I’d recommend trying to build some solid transforms to the adult spaces and pushing them to TemplateFlow. I believe Mathias Goncalves has been doing similar work with MNIInfant.
If you run fMRIPrep with MNIInfant as the output space, XCP-D should work, but I should warn you that some derivatives, like the parcellated time series, won’t be very high quality, since they rely on a pretty bad MNI152NLin6Asym-to-MNIInfant transform. The denoised data will be fine though, since it doesn’t rely on that transform.