Some algorithms (like period finding), use one or more measurement step. The post measurement state is then acted upon by another set of gates to complete the algorithm. If I imagine this as blackbox algorithm $f$ which takes $x$ as input and computes $f(x)$; then one can write a corresponding quantum circuit and use it as a subroutine in another quantum algorithm (like Grover's).
Now if one wants to implement this in a circuit, doesn't the measurement operation hiding inside $f$ create an issue of a possibly exponential number of measurements (like in Grover where we construct the full superposition)? I read somewhere that a CPTP operation can be embedded in a larger space as a unitary, but I recall it as an existence result. But even if that larger circuit has a construction, can't complexity/size of that circuit be much bigger?