As you have V7 and V8 to use you should look at the 'Multi Column Region' (MCR) functionality which should allow you to extract the data as you need to do.
There could be some problems if your report format is not consistent, as some are not, but in theory the MCR will work well and it takes a very unpleasantly formatted report to make life impossible for it.
The alternative might be to develop several models, one for each column of the report, and combine the outputs into a single file after (or during) export.
In effect this is what MCR will do in one pass.
To model your data using MCR pick one row and turn the MCR view on. Spoecify the number of columns and the width of the column format. You should see that the model is displaying fields in only the first column as you work on the definition. The detail template should have the field from the column and anything from the part fo the row to the LEFT of the first column - in this case Style and Colour fields - defined for the fields required.
In an append template select the column header descrption (assuming there is one) which should give you the Size value. The Column Header must be in the MCR for this to work but obviously using its own trap method to differentiate it from the detail template. If that is not possible for some reason ans alternative approach would be required.
Limit the columns by reference to where the MCR starts and ends in relation to other templates (which must exist before you can reference them of course!) This can be done during or after most of the definition has ben completed.
That should be all you need BUT the success rate does depend on how well structured and consistent the report is.
What the MCR does, in effect, is treat each column as a virtual page within the page but with each virtual page retaining whatever is available to the LEFT of the MCR.
I hope this helps. MCR can be very powerful but may appear a little strange at first I found. If your report is well structured you should be fine. If it is not well structured it might provide an interesting challenge.
Let us know if this works for you or, if it does not work, what problems you find.
If you have already looked at MCR and rejected it it would be interesting to understand the problem and and I qwould be happy to have a look at it and make some suggestions.