What is the role of kernel matrix in SVM?

Kernel matrix stores pairwise similarity between all training samples.

The kernel matrix K(i,j) = K(x_i, x_j) replaces explicit feature mapping. It enables computation in transformed feature space without ever computing coordinates directly. However, it requires O(n²) memory.