I'd suggest getting all this data into software. Then you'll get to the point where you can eliminate people.
(If you haven't chosen your genealogy software yet, you can busk it with Word for a while.)
So suppose you've started with 1851 censuses.
Transcribe ALL the likely looking ones. Start a separate record for each little family.
Now do the 1861 censuses. Assign each to one of your previous families.
And so on.
It may all fall into place from there.
If not, use FreeBMD to produce a near-as-dammit list of all the likely births for the people you are focussing on. The % figure in each result tells you what % of records from that quarter have been transcribed. If it's nearly 100% you can start making assumptions from completeness.
You can download your FreeBMD search results as a little file: assign these results to the family members you've already entered, crossing them off as you go.
Repeat with deaths (and, especially for women, marriages as that's how their names disappear).
You'll probably have some dubious cases - you can leave those unassigned or assign with a big ???.
Rinse and repeat with other sources.
Eventually you will come across something where it's clearly Mary SMITH b 1861/1862 dau of John & Helen who is the person in the record.
That means Mary Smith b 1860/1861 dau of Robert & Elizabeth must be the left over person.*
*Or there's an error in the records. So must...ish!