We employ so-called quantum kernel estimation to exploit complex quantum dynamics of solid-state nuclear magnetic resonance for machine learning. We propose to map an input to a feature space by input-dependent Hamiltonian evolution, and the kernel is estimated by the interference of the evolution. Simple machine learning tasks, namely one-dimensional regression tasks and two-dimensional classification tasks, are performed using proton spins which exhibit correlation over 10 spins. We also performed numerical simulations to evaluate the performance without the noise inevitable in the actual experiments. The performance of the trained model tends to increase with the longer evolution time, or equivalently, with a larger number of spins involved in the dynamics for certain tasks. This work presents a quantum machine learning experiment using one of the largest quantum systems to date.