Question

I know KDF (Key derivation function) are used to stretch user passwords, which are basically not suitable to be used as keys in cryptographic algorithms.

But what if I create a random key (random 32 bytes), do I still need to use KDF on it to ensure proper encryption?

Answer 1

A KDF is typically used for deriving cryptographic keys from things like passphrases, which as you correctly say are not suitable for direct use. But they are also used for deriving additional keys from a master key, which depending on your overall scheme, might be useful.

Suppose you used a key agreement protocol where both parties ended up with a random shared secret. You could use a KDF to derive a key for encryption, and one for message integrity (for example, an HMAC key).

From NIST SP800-108:

When parties share a secret symmetric key (e.g., upon a successful
execution of a key- establishment scheme as specified in 1 and
[2]), it is often the case that additional keys will be needed (e.g.
as described in [3]). Separate keys may be needed for different
cryptographic purposes – for example, one key may be required for an
encryption algorithm, while another key is intended for use by an
integrity protection algorithm, such as a message authentication
code. At other times, the distinct keys required by multiple entities
may be generated by a trusted party from a single master key. Key
derivation functions are used to derive such keys.

The short answer is, no, you don’t need to use a KDF, assuming your key generation is correct.