Two major areas of modern radio astronomy, namely, explosive astrophysical transient phenomena and observations of cosmological structures, are driving the design of aperture arrays towards large numbers of low-cost elements consisting of multiple spatial scales spanning the dimensions of individual elements, the size of stations (groupings of individual elements), and the spacing between stations. Such multi-scale, hierarchical aperture arrays require a combination of data processing architectures – pre-correlation beamformer, generic version of fast Fourier transform (FFT)-based direct imager, post-correlation beamformer, and post-correlation FFT imager – operating on different ranges of spatial scales to obtain optimal performance in imaging the entire field of view. Adopting a computational cost metric based on the number of floating point operations, its distribution over the dimensions of discovery space, namely, field of view, angular resolution, polarisation, frequency, and time is examined to determine the most efficient hybrid architectures over the parameter space of hierarchical aperture array layouts. Nominal parameters of specific upcoming and planned arrays – the SKA at low frequencies (SKA-low), SKA-low-core, a proposed long baseline extension to SKA-low (LAMBDA-I), Compact All-Sky Phased Array (CASPA), and a lunar array (FarView-core) – are used to determine the most optimal architecture hierarchy for each from a computational standpoint and provide a guide for designing hybrid architectures for multi-scale aperture arrays. For large, dense-packed layouts, a FFT-based direct imager is most efficient for most cadence intervals, and for other layouts that have relatively lesser number of elements or greater sparsity in distribution, the best architecture is more sensitive to the cadence interval, which in turn is determined by the science goals.