There exist public-key encryption (PKE) schemes where any well-formed ciphertext can be decrypted if the random coins used during encryption are known. For instance, in ElGamal encryption, the public key is $pk = (G, q, g, h = g^x)$ and the secret key is $x$. To encrypt a message $m \in G$, one selects a random $y \in \mathbb{Z}_q$ and produces the ciphertext $(c_0 = g^y, c_1 = m \cdot h^y)$. Here, the random value $y$ is the only randomness used, and knowing $y$ allows one to easily recover $m$.
Such schemes are used in the literature. For example, Koppula and Waters (see page 3) starts with a CPA-secure scheme that has this extraction property and uses it to construct a CCA-secure PKE scheme.
My question is: Are there provably CPA-secure PKE schemes that do not have such extraction property?