Dynamic functional connectivity (dFC) analyses of fMRI time-courses are typically performed using sliding-window based schemes. Such approaches not only inherently confine analysis to a single time-scale, but also do not generally lend themselves to accurate change-time estimates of the dynamically evolving graph topology. Change point detection methods on the other hand, offer the potential to overcome both limitations. However, the approaches employed so far in the dFC context are limited to detecting changes in linear relationships among time-courses corresponding to distinct regions of the brain. The present work puts forth a novel multi-kernel change point detection approach with the goal of capturing changes in the generally nonlinear relationships among time-courses, and thus in the topologies of the corresponding dynamically evolving FC graphs. The approach is tested on dynamic causal model (DCM) based synthetic resting-state fMRI data.