Thanks, Alfonso. Just so I'm clear then, what you're saying is that using this approach you have recommended is normalising the raw data as opposed to normalising the beta maps (SBC maps), because the beta maps derived from the connectivity analysis will be in native space?
So, then the workflow looks like this:

1.Define ROIs in subject space
2. Preprocess data outside on CONN in native space, and dont smooth. IMPORT into CONN and label as secondary dataset.
3. Import data from step 2 above and apply additional normalisation and smoothing into MNI space (call this primary dataset) 
4. In ROIs set up tab, select extract average BOLD signal from secondary dataset (subject-space data).

My point of confusion then, is if i set up 1st level analysis and do a seed-to-voxel with native-space ROIs as my seed and secondary dataset (native space) as the voxel, then the resulting SBC maps are still in native space and can't be normalised. So what is the purpose of creating another version of the data here that is normalised into MNI space? Or are you suggesting the better (recommended) approach is to do the 1st level analysis using the extracted time series from the ROIs in native space and the 'voxel' is the rest of the brain in 'MNI' space? 

Thank you again,
Bec