Independent component analysis has emerged as a promising approach for revealing structural relationships in multivariate dynamic systems, particularly in scenarios with limited knowledge of causal patterns. This article introduces a robust kernel-based maximum likelihood (KML) estimation method that accommodates the distributional characteristics of the structural sources of data variation. Our Monte Carlo study demonstrates the superior performance of the KML estimator compared to existing approaches for independence-based identification. Moreover, the proposed method enables partial identification and dimension reduction even in the presence of dependent shocks. We illustrate the benefits of our approach by applying it to the global oil market model of Kilian, highlighting its ability to capture unmodeled higher-order dependence between oil supply and speculative oil demand shocks.