I'd agree with the most of the previous posts, because there are so many
different ways of doing it that all of them will be right under some
circumstances!
If the project is for designing bespoke machinery, or maybe even some
consumer device where pretty much every part is unique to that design, then
using in-context relations or a skeleton part should cause few problems and
might save a stack of time.
However, if the project uses many existing components plus some new ones,
then in-context relations can be a minefield to pick your way through -
especially if the model was created by someone else!
I often use "empty parts" (i.e. just the standard planes + any useful
reference geometry) to fill in the gaps in the structure (and show up in a
provisional BOM) and to constrain other parts to. I'll then add the
geometry to them at the appropriate time.
One golden rule is that the "quickest" method to get something that looks
reasonable in the short term, is almost always the slowest method in the
longer term.
From past experience, the area where it is hardest to decide the approach is
when making "sales drawings" where you might need to quickly produce a
shaded image/eDrawing/drawing that looks vaguely believable to the customer.
How I tackle it depends on :-
a) whether I think the enquiry is highly speculative from a potential
customer or alternatively a "racing certainty" from an existing customer
b) whether it is basically a re-working of a previous design or
alternatively involves a lot of conceptual work.
At one end of the scale is a speculative enquiry involving conceptual design
work. Here I would tend to model parts that combine several/many real-world
parts - this saves on the overhead of naming/saving multiple files, adding
many components to an assembly, creating loads of mates, and obviates the
need for in-context relations. Obviously if parts have to move relative to
one another then they have to be separate.
The downside is that if the enquiry becomes an order, you'll have to remodel
it all as separate components. However, if you use multibody parts and are
careful to use reference geometry where possible, then you might be able to
salvage much of the geometry.
At the other end of the scale, are enquiries from existing customers that
are re-works of existing designs. Here I would make a copy of the top-level
assy and drawing and then selectively "save as" on any components/drawings
that needed to be new items.
If there is no well-constructed assembly as a starting point, then I'll use
a "bottom up" approach to add existing components to sub-assys which I'll
then add to a top level assy. Getting the structure right from the start
can save hours of heartache later.
I'll also use a top-down method to create the new parts.
John H